GL_ARB_enhanced_layouts, part 4: Numerical side of xfb_*: offset computation, size computation, alias detection, paddings, overflow, implicit strides, gl_Max* checks, etc.

git-svn-id: https://cvs.khronos.org/svn/repos/ogl/trunk/ecosystem/public/sdk/tools/glslang@25014 e7fa87d3-cd2b-0410-9028-fcbf551c1848

StandAlone/StandAlone.cpp

@@ -192,6 +192,8 @@ const char* DefaultConfig =
     "MaxFragmentAtomicCounterBuffers 1\n"
     "MaxCombinedAtomicCounterBuffers 1\n"
     "MaxAtomicCounterBufferSize 16384\n"
+    "MaxTransformFeedbackBuffers 4\n"
+    "MaxTransformFeedbackInterleavedComponents 64\n"
 
     "nonInductiveForLoops 1\n"
     "whileLoops 1\n"
@@ -390,6 +392,10 @@ void ProcessConfigFile()
             Resources.maxCombinedAtomicCounterBuffers = value;
         else if (strcmp(token, "MaxAtomicCounterBufferSize") == 0)
             Resources.maxAtomicCounterBufferSize = value;
+        else if (strcmp(token, "MaxTransformFeedbackBuffers") == 0)
+            Resources.maxTransformFeedbackBuffers = value;
+        else if (strcmp(token, "MaxTransformFeedbackInterleavedComponents") == 0)
+            Resources.maxTransformFeedbackInterleavedComponents = value;
 
         else if (strcmp(token, "nonInductiveForLoops") == 0)
             Resources.limits.nonInductiveForLoops = (value != 0);

Test/440.vert

@@ -68,31 +68,94 @@ out bblck1 {
 } bbinst1;
 
 out bblck2 {
-    layout(xfb_offset=16) vec4 bbv;
+    layout(xfb_offset=64) vec4 bbv;
 } bbinst2;
 
-layout(xfb_buffer = 3, xfb_stride = 16) out;
+layout(xfb_buffer = 3, xfb_stride = 64) out;  // default buffer is 3
 
 out bblck3 {
-    layout(xfb_offset=16) vec4 bbv;
+    layout(xfb_offset=16) vec4 bbv;  // in xfb_buffer 3
 } bbinst3;
 
 uniform ubblck3 {
-    layout(xfb_offset=16) vec4 bbv;  // ERROR
+    layout(xfb_offset=16) vec4 bbv;  // ERROR, not in a uniform
 } ubbinst3;
 
-layout(xfb_buffer=2, xfb_offset=32, xfb_stride=64) out vec4 bg;
+layout(xfb_buffer=2, xfb_offset=48, xfb_stride=80) out vec4 bg;
 layout(              xfb_offset=32, xfb_stride=64) out vec4 bh;
 
 layout(xfb_offset=48) out; // ERROR
 
-layout(xfb_stride=32, xfb_buffer=2, xfb_offset=16) out bblck4 {
+layout(xfb_stride=80, xfb_buffer=2, xfb_offset=16) out bblck4 {
     vec4 bbv1;
     vec4 bbv2;
 } bbinst4;
 
 out bblck5 {
     layout(xfb_offset=0) vec4 bbv1;
-    layout(xfb_stride=32, xfb_buffer=3, xfb_offset=16) vec4 bbv2;
+    layout(xfb_stride=64, xfb_buffer=3, xfb_offset=48) vec4 bbv2;
     layout(xfb_buffer=2) vec4 bbv3;                               // ERROR, wrong buffer
 } bbinst5;
+
+out layout(xfb_buffer=2) bblck6 {
+    layout(xfb_offset=0) vec4 bbv1;
+    layout(xfb_stride=64, xfb_buffer=3, xfb_offset=32) vec4 bbv2; // ERROR, overlap 32 from bh, and buffer contradiction
+    layout(xfb_buffer=2, xfb_offset=0) vec4 bbv3;                 // ERROR, overlap 0 from bbinst5
+    layout(xfb_buffer=2) vec4 bbv5;
+    layout(xfb_offset=24) float bbf6;                             // ERROR, overlap 24 from bbv1 in bbinst4
+} bbinst6;
+
+layout(xfb_stride=48) out;                   // ERROR, stride of buffer 3
+
+layout(xfb_buffer=1) out;  // default buffer is 1
+layout(xfb_offset=4) out float bj;
+layout(xfb_offset=0) out ivec2 bk;           // ERROR, overlap 4
+
+layout(xfb_buffer=3, xfb_stride=48) out;     // ERROR, stride of buffer 3 (default is now 3)
+layout(xfb_stride=48) out float bl;          // ERROR, stride of buffer 3
+
+layout(xfb_stride=48) out bblck7 {           // ERROR, stride of buffer 3
+    layout(xfb_stride=64) vec4 bbv1;
+    layout(xfb_stride=32) vec4 bbv2;         // ERROR, stride of buffer 3
+} bbinst7;
+
+struct S5 {
+    int i;    // 4 bytes plus 4 byte hole
+    double d; // 8 bytes
+    float f;  // 4 bytes
+};  // total size = 20
+
+struct T {
+    bool b;   // 4 plus 4 byte hole
+    S5 s;     // 20 
+    vec2 v2;  // 8
+};  // total size = 36
+
+out layout(xfb_buffer=0, xfb_offset=0, xfb_stride=92) bblck8 {  // ERROR, stride not multiple of 8
+    bool b;    // offset 0
+    T t;       // offset 8, size 40
+    int i;     // offset 40 + 4 = 48
+    mat3x3 m3; // offset 52
+    float f;   // offset 52 + 9*4 = 88
+    float g;   // ERROR, overflow stride
+} bbinst8;
+
+out layout(xfb_buffer=4) bblck9 {
+    layout(xfb_offset=1) bool b;     // ERROR
+    layout(xfb_offset=12) T t;       // ERROR
+    layout(xfb_offset=52) mat3x3 m3; // non-multiple of 8 okay
+    layout(xfb_offset=90) int i;     // ERROR
+    layout(xfb_offset=98) double d;  // ERROR
+    layout(xfb_offset=108) S s;      // non-multiple of 8 okay
+} bbinst9;
+
+layout(xfb_buffer=5, xfb_stride=6) out;     // link ERROR, stride not multiple of 4
+layout(xfb_offset=0) out float bm;
+
+layout(xfb_buffer=6, xfb_stride=2000) out;  // ERROR, stride too big
+
+out layout(xfb_buffer=7, xfb_offset=0) bblck10 {  // link ERROR, implicit stride too big
+    dmat4x4 m1;
+    dmat4x4 m2;
+    float f;
+} bbinst10;

Test/baseResults/440.vert.out

@@ -22,7 +22,27 @@ ERROR: 0:61: 'location' : cannot declare a default, use a full declaration
 ERROR: 0:81: 'xfb layout qualifier' : can only be used on an output 
 ERROR: 0:87: 'xfb_offset' : cannot declare a default, use a full declaration 
 ERROR: 0:97: 'xfb_buffer' : member cannot contradict block (or what block inherited from global) 
-ERROR: 22 compilation errors.  No code generated.
+ERROR: 0:102: 'xfb_buffer' : member cannot contradict block (or what block inherited from global) 
+ERROR: 0:102: 'xfb_offset' : overlapping offsets at offset 32 in buffer 3
+ERROR: 0:103: 'xfb_offset' : overlapping offsets at offset 0 in buffer 2
+ERROR: 0:105: 'xfb_offset' : overlapping offsets at offset 24 in buffer 2
+ERROR: 0:108: 'xfb_stride' : all stride settings must match for xfb buffer 15
+ERROR: 0:112: 'xfb_offset' : overlapping offsets at offset 4 in buffer 1
+ERROR: 0:114: 'xfb_stride' : all stride settings must match for xfb buffer 3
+ERROR: 0:115: 'xfb_stride' : all stride settings must match for xfb buffer 3
+ERROR: 0:119: 'xfb_stride' : all stride settings must match for xfb buffer 3
+ERROR: 0:117: 'xfb_stride' : all stride settings must match for xfb buffer 3
+ERROR: 0:138: 'xfb_offset' : overlapping offsets at offset 64 in buffer 0
+ERROR: 0:143: 'xfb_buffer' : buffer is too large: gl_MaxTransformFeedbackBuffers is 4
+ERROR: 0:144: 'xfb_offset' : must be a multiple of size of first component 
+ERROR: 0:145: 'xfb_offset' : type contains double; xfb_offset must be a multiple of 8 
+ERROR: 0:147: 'xfb_offset' : must be a multiple of size of first component 
+ERROR: 0:148: 'xfb_offset' : type contains double; xfb_offset must be a multiple of 8 
+ERROR: 0:152: 'xfb_buffer' : buffer is too large: gl_MaxTransformFeedbackBuffers is 4
+ERROR: 0:155: 'xfb_buffer' : buffer is too large: gl_MaxTransformFeedbackBuffers is 4
+ERROR: 0:155: 'xfb_stride' : 1/4 stride is too large: gl_MaxTransformFeedbackInterleavedComponents is 64
+ERROR: 0:157: 'xfb_buffer' : buffer is too large: gl_MaxTransformFeedbackBuffers is 4
+ERROR: 42 compilation errors.  No code generated.
 
 
 in xfb mode
@@ -60,13 +80,22 @@ ERROR: node is still EOpNull!
 0:?     'be' (layout(location=50 component=3 ) smooth out int)
 0:?     'bf' (layout(location=50 component=0 ) smooth out 3-component vector of float)
 0:?     'bbinst1' (out block{out 4-component vector of float bbv})
-0:?     'bbinst2' (out block{layout(xfb_buffer=0 xfb_offset=16 ) out 4-component vector of float bbv})
+0:?     'bbinst2' (out block{layout(xfb_buffer=0 xfb_offset=64 ) out 4-component vector of float bbv})
 0:?     'bbinst3' (out block{layout(xfb_buffer=3 xfb_offset=16 ) out 4-component vector of float bbv})
 0:?     'ubbinst3' (layout(column_major shared ) uniform block{layout(column_major shared xfb_offset=16 ) uniform 4-component vector of float bbv})
-0:?     'bg' (layout(xfb_buffer=2 xfb_offset=32 xfb_stride=64 ) smooth out 4-component vector of float)
+0:?     'bg' (layout(xfb_buffer=2 xfb_offset=48 xfb_stride=80 ) smooth out 4-component vector of float)
 0:?     'bh' (layout(xfb_buffer=3 xfb_offset=32 xfb_stride=64 ) smooth out 4-component vector of float)
-0:?     'bbinst4' (layout(xfb_buffer=2 xfb_offset=16 xfb_stride=32 ) out block{layout(xfb_buffer=2 xfb_offset=16 ) out 4-component vector of float bbv1, layout(xfb_buffer=2 xfb_offset=16 ) out 4-component vector of float bbv2})
+0:?     'bbinst4' (layout(xfb_stride=80 ) out block{layout(xfb_buffer=2 xfb_offset=16 ) out 4-component vector of float bbv1, layout(xfb_buffer=2 xfb_offset=32 ) out 4-component vector of float bbv2})
-0:?     'bbinst5' (out block{layout(xfb_buffer=3 xfb_offset=0 ) out 4-component vector of float bbv1, layout(xfb_buffer=3 xfb_offset=16 xfb_stride=32 ) out 4-component vector of float bbv2, out 4-component vector of float bbv3})
+0:?     'bbinst5' (out block{layout(xfb_buffer=3 xfb_offset=0 ) out 4-component vector of float bbv1, layout(xfb_buffer=3 xfb_offset=48 xfb_stride=64 ) out 4-component vector of float bbv2, out 4-component vector of float bbv3})
+0:?     'bbinst6' (out block{layout(xfb_buffer=2 xfb_offset=0 ) out 4-component vector of float bbv1, layout(xfb_buffer=3 xfb_offset=32 xfb_stride=64 ) out 4-component vector of float bbv2, layout(xfb_buffer=2 xfb_offset=0 ) out 4-component vector of float bbv3, out 4-component vector of float bbv5, layout(xfb_buffer=2 xfb_offset=24 ) out float bbf6})
+0:?     'bj' (layout(xfb_buffer=1 xfb_offset=4 ) smooth out float)
+0:?     'bk' (layout(xfb_buffer=1 xfb_offset=0 ) smooth out 2-component vector of int)
+0:?     'bl' (layout(xfb_stride=48 ) smooth out float)
+0:?     'bbinst7' (layout(xfb_stride=48 ) out block{layout(xfb_stride=64 ) out 4-component vector of float bbv1, layout(xfb_stride=32 ) out 4-component vector of float bbv2})
+0:?     'bbinst8' (layout(xfb_stride=92 ) out block{layout(xfb_buffer=0 xfb_offset=0 ) out bool b, layout(xfb_buffer=0 xfb_offset=8 ) out structure{bool b, structure{int i, double d, float f} s, 2-component vector of float v2} t, layout(xfb_buffer=0 xfb_offset=48 ) out int i, layout(xfb_buffer=0 xfb_offset=52 ) out 3X3 matrix of float m3, layout(xfb_buffer=0 xfb_offset=88 ) out float f, layout(xfb_buffer=0 xfb_offset=92 ) out float g})
+0:?     'bbinst9' (out block{layout(xfb_buffer=4 xfb_offset=1 ) out bool b, layout(xfb_buffer=4 xfb_offset=12 ) out structure{bool b, structure{int i, double d, float f} s, 2-component vector of float v2} t, layout(xfb_buffer=4 xfb_offset=52 ) out 3X3 matrix of float m3, layout(xfb_buffer=4 xfb_offset=90 ) out int i, layout(xfb_buffer=4 xfb_offset=98 ) out double d, layout(xfb_buffer=4 xfb_offset=108 ) out structure{int a} s})
+0:?     'bm' (layout(xfb_buffer=5 xfb_offset=0 ) smooth out float)
+0:?     'bbinst10' (out block{layout(xfb_buffer=7 xfb_offset=0 ) out 4X4 matrix of double m1, layout(xfb_buffer=7 xfb_offset=128 ) out 4X4 matrix of double m2, layout(xfb_buffer=7 xfb_offset=256 ) out float f})
 0:?     'gl_VertexID' (gl_VertexId int)
 0:?     'gl_InstanceID' (gl_InstanceId int)
 
@@ -74,6 +103,14 @@ ERROR: node is still EOpNull!
 Linked vertex stage:
 
 ERROR: Linking vertex stage: Missing entry point: Each stage requires one "void main()" entry point
+ERROR: Linking vertex stage: xfb_stride is too small to hold all buffer entries:
+ERROR:     xfb_buffer 0, xfb_stride 92, minimum stride needed: 96
+ERROR: Linking vertex stage: xfb_stride must be multiple of 8 for buffer holding a double:
+ERROR:     xfb_buffer 0, xfb_stride 92
+ERROR: Linking vertex stage: xfb_stride must be multiple of 4:
+ERROR:     xfb_buffer 5, xfb_stride 6
+ERROR: Linking vertex stage: xfb_stride is too large:
+ERROR:     xfb_buffer 7, components (1/4 stride) needed are 66, gl_MaxTransformFeedbackInterleavedComponents is 64
 
 in xfb mode
 

Test/baseResults/test.conf

@@ -75,6 +75,8 @@ MaxGeometryAtomicCounterBuffers 0
 MaxFragmentAtomicCounterBuffers 1
 MaxCombinedAtomicCounterBuffers 1
 MaxAtomicCounterBufferSize 16384
+MaxTransformFeedbackBuffers 4
+MaxTransformFeedbackInterleavedComponents 64
 nonInductiveForLoops 1
 whileLoops 1
 doWhileLoops 1

glslang/Include/Common.h

@@ -185,6 +185,20 @@ typedef TMap<TString, TString>::tAllocator TPragmaTableAllocator;
 
 const int GlslangMaxTokenLength = 1024;
 
+// Round number up to a multiple of the given powerOf2, which is not
+// a power, just a number that must be a power of 2.
+template <class T> void RoundToPow2(T& number, int powerOf2)
+{
+    assert((powerOf2 & (powerOf2 - 1)) == 0);
+    number = (number + powerOf2 - 1) & ~(powerOf2 - 1);
+}
+
+template <class T> bool IsMultipleOfPow2(T number, int powerOf2)
+{
+    assert((powerOf2 & (powerOf2 - 1)) == 0);
+    return ! (number & (powerOf2 - 1));
+}
+
 } // end namespace glslang
 
 #endif // _COMMON_INCLUDED_

glslang/Include/ResourceLimits.h

@@ -127,6 +127,8 @@ struct TBuiltInResource {
     int maxFragmentAtomicCounterBuffers;
     int maxCombinedAtomicCounterBuffers;
     int maxAtomicCounterBufferSize;
+    int maxTransformFeedbackBuffers;
+    int maxTransformFeedbackInterleavedComponents;
 
     TLimits limits;
 };

glslang/Include/Types.h

@@ -373,24 +373,32 @@ public:
                hasStream() ||
                hasXfb();
     }
-    TLayoutMatrix  layoutMatrix                     : 3;
+    TLayoutMatrix  layoutMatrix  : 3;
-    TLayoutPacking layoutPacking                    : 4;
+    TLayoutPacking layoutPacking : 4;
     int layoutOffset;
     int layoutAlign;
-                 unsigned int layoutLocation        : 7;
+
-    static const unsigned int layoutLocationEnd =  0x3F;
+                 unsigned int layoutLocation         : 7;
-                 unsigned int layoutComponent       : 3;
+    static const unsigned int layoutLocationEnd =   0x3F;
-    static const unsigned int layoutComponentEnd =    4;
+
-                 unsigned int layoutBinding         : 8;
+                 unsigned int layoutComponent        : 3;
-    static const unsigned int layoutBindingEnd =   0xFF;
+    static const unsigned int layoutComponentEnd =     4;
-                 unsigned int layoutStream          : 8;
+
-    static const unsigned int layoutStreamEnd =    0xFF;    
+                 unsigned int layoutBinding          : 8;
-                 unsigned int layoutXfbBuffer       : 4;
+    static const unsigned int layoutBindingEnd =    0xFF;
-    static const unsigned int layoutXfbBufferEnd =  0xF;
+
-                 unsigned int layoutXfbStride       : 8;
+                 unsigned int layoutStream           : 8;
-    static const unsigned int layoutXfbStrideEnd = 0xFF;
+    static const unsigned int layoutStreamEnd =     0xFF;
-                 unsigned int layoutXfbOffset       : 8;
+
-    static const unsigned int layoutXfbOffsetEnd = 0xFF;
+                 unsigned int layoutXfbBuffer        : 4;
+    static const unsigned int layoutXfbBufferEnd =   0xF;
+
+                 unsigned int layoutXfbStride       : 10;
+    static const unsigned int layoutXfbStrideEnd = 0x3FF;
+
+                 unsigned int layoutXfbOffset       : 10;
+    static const unsigned int layoutXfbOffsetEnd = 0x3FF;
+
     bool hasUniformLayout() const
     {
         return layoutMatrix  != ElmNone ||
@@ -805,6 +813,20 @@ public:
     virtual bool isArray()  const { return arraySizes != 0; }
     virtual bool isStruct() const { return structure != 0; }
 
+    // Recursively checks if the type contains the given basic type
+    virtual bool containsBasicType(TBasicType checkType) const
+    {
+        if (basicType == checkType)
+            return true;
+        if (! structure)
+            return false;
+        for (unsigned int i = 0; i < structure->size(); ++i) {
+            if ((*structure)[i].type->containsBasicType(checkType))
+                return true;
+        }
+        return false;
+    }
+
     // Recursively check the structure for any arrays, needed for some error checks
     virtual bool containsArray() const
     {

glslang/Include/revision.h

@@ -9,5 +9,5 @@
 // source have to figure out how to create revision.h just to get a build
 // going.  However, if it is not updated, it can be a version behind.
 
-#define GLSLANG_REVISION "24964"
+#define GLSLANG_REVISION "24977"
-#define GLSLANG_DATE     "2014/01/22 17:35:24"
+#define GLSLANG_DATE     "2014/01/23 14:40:33"

glslang/MachineIndependent/ParseHelper.cpp

@@ -104,11 +104,16 @@ TParseContext::TParseContext(TSymbolTable& symt, TIntermediate& interm, bool pb,
     globalInputDefaults.clear();
 
     globalOutputDefaults.clear();
+
+    // "Shaders in the transform 
+    // feedback capturing mode have an initial global default of
+    //     layout(xfb_buffer = 0) out;"
     if (language == EShLangVertex ||
         language == EShLangTessControl ||
         language == EShLangTessEvaluation ||
         language == EShLangGeometry)
         globalOutputDefaults.layoutXfbBuffer = 0;
+
     if (language == EShLangGeometry)
         globalOutputDefaults.layoutStream = 0;
 }
@@ -123,6 +128,8 @@ void TParseContext::setLimits(const TBuiltInResource& r)
                      ! limits.generalUniformIndexing ||
                      ! limits.generalVariableIndexing ||
                      ! limits.generalVaryingIndexing;
+
+    intermediate.setLimits(resources);
 }
 
 //
@@ -2872,27 +2879,39 @@ void TParseContext::setLayoutQualifier(TSourceLoc loc, TPublicType& publicType,
             publicType.qualifier.layoutComponent = value;
         return;
     } else if (id.compare(0, 4, "xfb_") == 0) {
+        // "Any shader making any static use (after preprocessing) of any of these 
+        // *xfb_* qualifiers will cause the shader to be in a transform feedback 
+        // capturing mode and hence responsible for describing the transform feedback 
+        // setup."
         intermediate.setXfbMode();
         const char* feature = "transform feedback qualifier";
         requireStage(loc, (EShLanguageMask)(EShLangVertexMask | EShLangGeometryMask | EShLangTessControlMask | EShLangTessEvaluationMask), feature);
         requireProfile(loc, ECoreProfile | ECompatibilityProfile, feature);
         profileRequires(loc, ECoreProfile | ECompatibilityProfile, 440, GL_ARB_enhanced_layouts, feature);
         if (id == "xfb_buffer") {
-            if (value >= TQualifier::layoutXfbBufferEnd)   // TODO: 4.4 enhanced layouts: also check against gl_MaxTransformFeedbackBuffers
+            // "It is a compile-time error to specify an *xfb_buffer* that is greater than
-                error(loc, "buffer is too large", id.c_str(), "");
+            // the implementation-dependent constant gl_MaxTransformFeedbackBuffers."
+            if (value >= resources.maxTransformFeedbackBuffers)
+                error(loc, "buffer is too large:", id.c_str(), "gl_MaxTransformFeedbackBuffers is %d", resources.maxTransformFeedbackBuffers);                
+            if (value >= TQualifier::layoutXfbBufferEnd)
+                error(loc, "buffer is too large:", id.c_str(), "internal max is %d", TQualifier::layoutXfbBufferEnd-1);
             else
                 publicType.qualifier.layoutXfbBuffer = value;
             return;
         } else if (id == "xfb_offset") {
-            if (value >= TQualifier::layoutXfbOffsetEnd)   // TODO: 4.4 enhanced layouts: also check against gl_MaxTransformFeedbackInterleavedComponents
+            if (value >= TQualifier::layoutXfbOffsetEnd)
-                error(loc, "offset is too large", id.c_str(), "");
+                error(loc, "offset is too large:", id.c_str(), "internal max is %d", TQualifier::layoutXfbOffsetEnd-1);
             else
                 publicType.qualifier.layoutXfbOffset = value;
             return;
         } else if (id == "xfb_stride") {
-            if (value >= TQualifier::layoutXfbStrideEnd)    // TODO: 4.4 enhanced layouts: also check against 4*gl_MaxTransformFeedbackInterleavedComponents
+            // "The resulting stride (implicit or explicit), when divided by 4, must be less than or equal to the 
-                error(loc, "stride is too large", id.c_str(), "");
+            // implementation-dependent constant gl_MaxTransformFeedbackInterleavedComponents."
-            else
+            if (value > 4 * resources.maxTransformFeedbackInterleavedComponents)
+                error(loc, "1/4 stride is too large:", id.c_str(), "gl_MaxTransformFeedbackInterleavedComponents is %d", resources.maxTransformFeedbackInterleavedComponents);
+            else if (value >= TQualifier::layoutXfbStrideEnd)
+                error(loc, "stride is too large:", id.c_str(), "internal max is %d", TQualifier::layoutXfbStrideEnd-1);
+            if (value < TQualifier::layoutXfbStrideEnd)
                 publicType.qualifier.layoutXfbStride = value;
             return;
         }
@@ -2956,8 +2975,6 @@ void TParseContext::mergeObjectLayoutQualifiers(TSourceLoc loc, TQualifier& dst,
 
     if (src.hasXfbBuffer())
         dst.layoutXfbBuffer = src.layoutXfbBuffer;
-    if (src.hasXfbOffset())
-        dst.layoutXfbOffset = src.layoutXfbOffset;
 
     if (! inheritOnly) {
         if (src.layoutLocation != TQualifier::layoutLocationEnd)
@@ -2975,6 +2992,8 @@ void TParseContext::mergeObjectLayoutQualifiers(TSourceLoc loc, TQualifier& dst,
 
         if (src.hasXfbStride())
             dst.layoutXfbStride = src.layoutXfbStride;
+        if (src.hasXfbOffset())
+            dst.layoutXfbOffset = src.layoutXfbOffset;
     }
 }
 
@@ -3069,6 +3088,25 @@ void TParseContext::layoutTypeCheck(TSourceLoc loc, const TType& type)
             error(loc, "fragment outputs sharing the same location must be the same basic type", "location", "%d", repeated);
     }
 
+    if (qualifier.hasXfbOffset() && qualifier.hasXfbBuffer()) {
+        int repeated = intermediate.addXfbBufferOffset(type);
+        if (repeated >= 0)
+            error(loc, "overlapping offsets at", "xfb_offset", "offset %d in buffer %d", repeated, qualifier.layoutXfbBuffer);
+
+        // "The offset must be a multiple of the size of the first component of the first 
+        // qualified variable or block member, or a compile-time error results. Further, if applied to an aggregate 
+        // containing a double, the offset must also be a multiple of 8..."
+        if (type.containsBasicType(EbtDouble) && ! IsMultipleOfPow2(qualifier.layoutXfbOffset, 8))
+            error(loc, "type contains double; xfb_offset must be a multiple of 8", "xfb_offset", "");
+        else if (! IsMultipleOfPow2(qualifier.layoutXfbOffset, 4))
+            error(loc, "must be a multiple of size of first component", "xfb_offset", "");
+    }
+
+    if (qualifier.hasXfbStride() && qualifier.hasXfbBuffer()) {
+        if (! intermediate.setXfbBufferStride(qualifier.layoutXfbBuffer, qualifier.layoutXfbStride))
+            error(loc, "all stride settings must match for xfb buffer", "xfb_stride", "%d", qualifier.layoutXfbBuffer);
+    }
+
     if (qualifier.hasBinding()) {
         // Binding checking, from the spec:
         //
@@ -3844,10 +3882,16 @@ void TParseContext::declareBlock(TSourceLoc loc, TTypeList& typeList, const TStr
             if (defaultQualification.layoutStream != memberQualifier.layoutStream)
                 error(memberLoc, "member cannot contradict block", "stream", "");
         }
+
+        // "This includes a block's inheritance of the 
+        // current global default buffer, a block member's inheritance of the block's 
+        // buffer, and the requirement that any *xfb_buffer* declared on a block 
+        // member must match the buffer inherited from the block."
         if (memberQualifier.hasXfbBuffer()) {
             if (defaultQualification.layoutXfbBuffer != memberQualifier.layoutXfbBuffer)
                 error(memberLoc, "member cannot contradict block (or what block inherited from global)", "xfb_buffer", "");
         }
+
         if (memberQualifier.layoutPacking != ElpNone)
             error(memberLoc, "member of block cannot have a packing layout qualifier", typeList[member].type->getFieldName().c_str(), "");
         if (memberQualifier.hasLocation()) {
@@ -3869,7 +3913,10 @@ void TParseContext::declareBlock(TSourceLoc loc, TTypeList& typeList, const TStr
         mergeQualifiers(memberLoc, newMemberQualification, memberQualifier, false);
         memberQualifier = newMemberQualification;
     }
+
+    // Process the members
     fixBlockLocations(loc, currentBlockQualifier, typeList, memberWithLocation, memberWithoutLocation);
+    fixBlockXfbOffsets(loc, currentBlockQualifier, typeList);
     for (unsigned int member = 0; member < typeList.size(); ++member)
         layoutTypeCheck(typeList[member].loc, *typeList[member].type);
 
@@ -3983,6 +4030,37 @@ void TParseContext::fixBlockLocations(TSourceLoc loc, TQualifier& qualifier, TTy
     }
 }
 
+void TParseContext::fixBlockXfbOffsets(TSourceLoc loc, TQualifier& qualifier, TTypeList& typeList)
+{
+    // "If a block is qualified with xfb_offset, all its 
+    // members are assigned transform feedback buffer offsets. If a block is not qualified with xfb_offset, any 
+    // members of that block not qualified with an xfb_offsetwill not be assigned transform feedback buffer 
+    // offsets."
+
+    if (! currentBlockQualifier.hasXfbBuffer() || ! currentBlockQualifier.hasXfbOffset())
+        return;
+
+    int nextOffset = currentBlockQualifier.layoutXfbOffset;
+    for (unsigned int member = 0; member < typeList.size(); ++member) {
+        TQualifier& memberQualifier = typeList[member].type->getQualifier();
+        bool containsDouble = false;
+        int memberSize = intermediate.computeTypeXfbSize(*typeList[member].type, containsDouble);
+        // see if we need to auto-assign an offset to this member
+        if (! memberQualifier.hasXfbOffset()) {
+            // "if applied to an aggregate containing a double, the offset must also be a multiple of 8"
+            if (containsDouble)
+                RoundToPow2(nextOffset, 8);
+            memberQualifier.layoutXfbOffset = nextOffset;
+        } else
+            nextOffset = memberQualifier.layoutXfbOffset;
+        nextOffset += memberSize;
+    }
+
+    // The above gave all block members an offset, so we can take it off the block now,
+    // which will avoid double counting the offset usage.
+    qualifier.layoutXfbOffset = TQualifier::layoutXfbOffsetEnd;
+}
+
 // For an identifier that is already declared, add more qualification to it.
 void TParseContext::addQualifierToExisting(TSourceLoc loc, TQualifier qualifier, const TString& identifier)
 {
@@ -4150,6 +4228,10 @@ void TParseContext::updateStandaloneQualifierDefaults(TSourceLoc loc, const TPub
             globalOutputDefaults.layoutStream = qualifier.layoutStream;
         if (qualifier.hasXfbBuffer())
             globalOutputDefaults.layoutXfbBuffer = qualifier.layoutXfbBuffer;
+        if (globalOutputDefaults.hasXfbBuffer() && qualifier.hasXfbStride()) {
+            if (! intermediate.setXfbBufferStride(globalOutputDefaults.layoutXfbBuffer, qualifier.layoutXfbStride))
+                error(loc, "all stride settings must match for xfb buffer", "xfb_stride", "%d", qualifier.layoutXfbBuffer);
+        }
         break;
     default:
         error(loc, "default qualifier requires 'uniform', 'buffer', 'in', or 'out' storage qualification", "", "");

glslang/MachineIndependent/ParseHelper.h

@@ -165,6 +165,7 @@ public:
     TIntermTyped* constructBuiltIn(const TType&, TOperator, TIntermNode*, TSourceLoc, bool subset);
     void declareBlock(TSourceLoc, TTypeList& typeList, const TString* instanceName = 0, TArraySizes* arraySizes = 0);
     void fixBlockLocations(TSourceLoc, TQualifier&, TTypeList&, bool memberWithLocation, bool memberWithoutLocation);
+    void fixBlockXfbOffsets(TSourceLoc, TQualifier&, TTypeList&);
     void addQualifierToExisting(TSourceLoc, TQualifier, const TString& identifier);
     void addQualifierToExisting(TSourceLoc, TQualifier, TIdentifierList&);
     void invariantCheck(TSourceLoc, const TType&, const TString& identifier);

glslang/MachineIndependent/linkValidate.cpp

@@ -112,6 +112,16 @@ void TIntermediate::merge(TInfoSink& infoSink, TIntermediate& unit)
 
     if (unit.xfbMode)
         xfbMode = true;
+    for (size_t b = 0; b < xfbBuffers.size(); ++b) {
+        if (xfbBuffers[b].stride == TQualifier::layoutXfbStrideEnd)
+            xfbBuffers[b].stride = unit.xfbBuffers[b].stride;
+        else if (xfbBuffers[b].stride != unit.xfbBuffers[b].stride)
+            error(infoSink, "Contradictory xfb_stride");
+        xfbBuffers[b].implicitStride = std::max(xfbBuffers[b].implicitStride, unit.xfbBuffers[b].implicitStride);
+        if (unit.xfbBuffers[b].containsDouble)
+            xfbBuffers[b].containsDouble = true;
+        // TODO: 4.4 link: enhanced layouts: compare ranges
+    }
 
     if (unit.treeRoot == 0)
         return;
@@ -305,6 +315,44 @@ void TIntermediate::finalCheck(TInfoSink& infoSink)
     if (inIoAccessed("gl_FragColor") && inIoAccessed("gl_FragData"))
         error(infoSink, "Cannot use both gl_FragColor and gl_FragData");
 
+    for (size_t b = 0; b < xfbBuffers.size(); ++b) {
+        if (xfbBuffers[b].containsDouble)
+            RoundToPow2(xfbBuffers[b].implicitStride, 8);
+
+        // "It is a compile-time or link-time error to have 
+        // any xfb_offset that overflows xfb_stride, whether stated on declarations before or after the xfb_stride, or
+        // in different compilation units. While xfb_stridecan be declared multiple times for the same buffer, it is a
+        // compile-time or link-time error to have different values specified for the stride for the same buffer."
+        if (xfbBuffers[b].stride != TQualifier::layoutXfbStrideEnd && xfbBuffers[b].implicitStride > xfbBuffers[b].stride) {
+            error(infoSink, "xfb_stride is too small to hold all buffer entries:");
+            infoSink.info.prefix(EPrefixError);
+            infoSink.info << "    xfb_buffer " << b << ", xfb_stride " << xfbBuffers[b].stride << ", minimum stride needed: " << xfbBuffers[b].implicitStride << "\n";
+        }
+        if (xfbBuffers[b].stride == TQualifier::layoutXfbStrideEnd)
+            xfbBuffers[b].stride = xfbBuffers[b].implicitStride;
+
+        // "If the buffer is capturing any 
+        // outputs with double-precision components, the stride must be a multiple of 8, otherwise it must be a 
+        // multiple of 4, or a compile-time or link-time error results."
+        if (xfbBuffers[b].containsDouble && ! IsMultipleOfPow2(xfbBuffers[b].stride, 8)) {
+            error(infoSink, "xfb_stride must be multiple of 8 for buffer holding a double:");
+            infoSink.info.prefix(EPrefixError);
+            infoSink.info << "    xfb_buffer " << b << ", xfb_stride " << xfbBuffers[b].stride << "\n";
+        } else if (! IsMultipleOfPow2(xfbBuffers[b].stride, 4)) {
+            error(infoSink, "xfb_stride must be multiple of 4:");
+            infoSink.info.prefix(EPrefixError);
+            infoSink.info << "    xfb_buffer " << b << ", xfb_stride " << xfbBuffers[b].stride << "\n";
+        }
+
+        // "The resulting stride (implicit or explicit), when divided by 4, must be less than or equal to the 
+        // implementation-dependent constant gl_MaxTransformFeedbackInterleavedComponents."
+        if (xfbBuffers[b].stride > (unsigned int)(4 * resources.maxTransformFeedbackInterleavedComponents)) {
+            error(infoSink, "xfb_stride is too large:");
+            infoSink.info.prefix(EPrefixError);
+            infoSink.info << "    xfb_buffer " << b << ", components (1/4 stride) needed are " << xfbBuffers[b].stride/4 << ", gl_MaxTransformFeedbackInterleavedComponents is " << resources.maxTransformFeedbackInterleavedComponents << "\n";
+        }
+    }
+
     switch (language) {
     case EShLangVertex:
         break;
@@ -510,6 +558,7 @@ int TIntermediate::addUsedLocation(const TQualifier& qualifier, const TType& typ
         else
             size = 1;
     } else {
+        // Strip off the outer array dimension for those having an extra one.
         if (type.isArray() && ! qualifier.patch &&
             (language == EShLangGeometry && qualifier.isPipeInput()) ||
             language == EShLangTessControl                           ||
@@ -520,36 +569,31 @@ int TIntermediate::addUsedLocation(const TQualifier& qualifier, const TType& typ
             size = computeTypeLocationSize(type);
     }
 
-    TRange locationRange = { qualifier.layoutLocation, qualifier.layoutLocation + size - 1 };
+    TRange locationRange(qualifier.layoutLocation, qualifier.layoutLocation + size - 1);
-    TRange componentRange = { 0, 3 };
+    TRange componentRange(0, 3);
     if (qualifier.layoutComponent != TQualifier::layoutComponentEnd) {
         componentRange.start = qualifier.layoutComponent;
         componentRange.last = componentRange.start + type.getVectorSize() - 1;
     }
+    TIoRange range(locationRange, componentRange, type.getBasicType());
 
     // check for collisions, except for vertex inputs on desktop
     if (! (profile != EEsProfile && language == EShLangVertex && qualifier.isPipeInput())) {
         for (size_t r = 0; r < usedIo[set].size(); ++r) {
-            if (locationRange.last  >= usedIo[set][r].location.start &&
+            if (range.overlap(usedIo[set][r])) {
-                locationRange.start <= usedIo[set][r].location.last &&
-                componentRange.last  >= usedIo[set][r].component.start &&
-                componentRange.start <= usedIo[set][r].component.last) {
                 // there is a collision; pick one
                 return std::max(locationRange.start, usedIo[set][r].location.start);
-            } else if (locationRange.last  >= usedIo[set][r].location.start &&
+            } else if (locationRange.overlap(usedIo[set][r].location) && type.getBasicType() != usedIo[set][r].basicType) {
-                       locationRange.start <= usedIo[set][r].location.last &&
+                // aliased-type mismatch
-                       type.getBasicType() != usedIo[set][r].basicType) {
                 typeCollision = true;
                 return std::max(locationRange.start, usedIo[set][r].location.start);
             }
         }
     }
 
-    TIoRange range = { locationRange, componentRange, type.getBasicType() };
-
     usedIo[set].push_back(range);
 
-    return -1;
+    return -1; // no collision
 }
 
 // Recursively figure out how many locations are used up by an input or output type.
@@ -605,4 +649,95 @@ int TIntermediate::computeTypeLocationSize(const TType& type)
     return 1;
 }
 
+// Accumulate xfb buffer ranges and check for collisions as the accumulation is done.
+//
+// Returns < 0 if no collision, >= 0 if collision and the value returned is a colliding value.
+//
+int TIntermediate::addXfbBufferOffset(const TType& type)
+{
+    const TQualifier& qualifier = type.getQualifier();
+
+    assert(qualifier.hasXfbOffset() && qualifier.hasXfbBuffer());
+    TXfbBuffer& buffer = xfbBuffers[qualifier.layoutXfbBuffer];
+
+    // compute the range
+    unsigned int size = computeTypeXfbSize(type, buffer.containsDouble);
+    buffer.implicitStride = std::max(buffer.implicitStride, qualifier.layoutXfbOffset + size);
+    TRange range(qualifier.layoutXfbOffset, qualifier.layoutXfbOffset + size - 1);
+
+    // check for collisions
+    for (size_t r = 0; r < buffer.ranges.size(); ++r) {
+        if (range.overlap(buffer.ranges[r])) {
+            // there is a collision; pick an example to return
+            return std::max(range.start, buffer.ranges[r].start);
+        }
+    }
+
+    buffer.ranges.push_back(range);
+
+    return -1;  // no collision
+}
+
+// Recursively figure out how many bytes of xfb buffer are used by the given type.
+// Return the size of type, in bytes.
+// Sets containsDouble to true if the type contains a double.
+// N.B. Caller must set containsDouble to false before calling.
+unsigned int TIntermediate::computeTypeXfbSize(const TType& type, bool& containsDouble) const
+{
+    // "...if applied to an aggregate containing a double, the offset must also be a multiple of 8, 
+    // and the space taken in the buffer will be a multiple of 8.
+    // ...within the qualified entity, subsequent components are each 
+    // assigned, in order, to the next available offset aligned to a multiple of
+    // that component's size.  Aggregate types are flattened down to the component
+    // level to get this sequence of components."
+
+    if (type.isArray()) {
+        assert(type.getArraySize() > 0);
+        TType elementType(type, 0);
+        return type.getArraySize() * computeTypeXfbSize(elementType, containsDouble);
+    }
+
+    if (type.isStruct()) {
+        unsigned int size = 0;
+        bool structContainsDouble = false;
+        for (size_t member = 0; member < type.getStruct()->size(); ++member) {
+            TType memberType(type, member);
+            // "... if applied to 
+            // an aggregate containing a double, the offset must also be a multiple of 8, 
+            // and the space taken in the buffer will be a multiple of 8."
+            bool memberContainsDouble = false;
+            int memberSize = computeTypeXfbSize(memberType, memberContainsDouble);
+            if (memberContainsDouble) {
+                structContainsDouble = true;
+                RoundToPow2(size, 8);
+            }
+            size += memberSize;
+        }
+
+        if (structContainsDouble) {
+            containsDouble = true;
+            RoundToPow2(size, 8);
+        }
+        return size;
+    }
+
+    int numComponents;
+    if (type.isScalar())
+        numComponents = 1;
+    else if (type.isVector())
+        numComponents = type.getVectorSize();
+    else if (type.isMatrix())
+        numComponents = type.getMatrixCols() * type.getMatrixRows();
+    else {
+        assert(0);
+        numComponents = 1;
+    }
+
+    if (type.getBasicType() == EbtDouble) {
+        containsDouble = true;
+        return 8 * numComponents;
+    } else
+        return 4 * numComponents;
+}
+
 } // end namespace glslang

glslang/MachineIndependent/localintermediate.h

@@ -51,6 +51,55 @@ struct TVectorFields {
     int num;
 };
 
+//
+// Some helper structures for TIntermediate.  Their contents are encapsulated
+// by TIntermediate.
+//
+
+// Used for detecting recursion:  A "call" is a pair: <caller, callee>.
+struct TCall {
+    TCall(const TString& pCaller, const TString& pCallee) : caller(pCaller), callee(pCallee) { }
+    TString caller;
+    TString callee;
+    bool visited;
+    bool currentPath;
+    bool errorGiven;
+};
+
+// A generic 1-D range.
+struct TRange {
+    TRange(int start, int last) : start(start), last(last) { }
+    bool overlap(const TRange& rhs) const 
+    {
+        return last >= rhs.start && start <= rhs.last;
+    }
+    int start;
+    int last;
+};
+
+// A *location* range is a 2-D rectangle; the set of (location, component) pairs all lying
+// both within the location range and the component range.  Locations don't alias unless
+// both dimensions of their range overlap.
+struct TIoRange {
+    TIoRange(TRange location, TRange component, TBasicType basicType) : location(location), component(component), basicType(basicType) { }
+    bool overlap(const TIoRange& rhs) const
+    {
+        return location.overlap(rhs.location) && component.overlap(rhs.component);
+    }
+    TRange location;
+    TRange component;
+    TBasicType basicType;
+};
+
+// Things that need to be tracked per xfb buffer.
+struct TXfbBuffer {
+    TXfbBuffer() : stride(TQualifier::layoutXfbStrideEnd), implicitStride(0), containsDouble(false) { }
+    std::vector<TRange> ranges;  // byte offsets that have already been assigned
+    unsigned int stride;
+    unsigned int implicitStride;
+    bool containsDouble;
+};
+
 class TSymbolTable;
 class TSymbol;
 
@@ -62,7 +111,12 @@ public:
     explicit TIntermediate(EShLanguage l, int v = 0, EProfile p = ENoProfile) : language(l), treeRoot(0), profile(p), version(v), 
         numMains(0), numErrors(0), recursive(false),
         invocations(0), vertices(0), inputPrimitive(ElgNone), outputPrimitive(ElgNone), pixelCenterInteger(false), originUpperLeft(false),
-        vertexSpacing(EvsNone), vertexOrder(EvoNone), pointMode(false), xfbMode(false) { }
+        vertexSpacing(EvsNone), vertexOrder(EvoNone), pointMode(false), xfbMode(false)
+    {
+        xfbBuffers.resize(TQualifier::layoutXfbBufferEnd);
+    }
+    void setLimits(const TBuiltInResource& r) { resources = r; }
+
     bool postProcess(TIntermNode*, EShLanguage);
     void output(TInfoSink&, bool tree);
 	void removeTree();
@@ -176,6 +230,16 @@ public:
     int addUsedLocation(const TQualifier&, const TType&, bool& typeCollision);
     int computeTypeLocationSize(const TType&);
 
+    bool setXfbBufferStride(int buffer, int stride)
+    {
+        if (xfbBuffers[buffer].stride != TQualifier::layoutXfbStrideEnd)
+            return xfbBuffers[buffer].stride == stride;
+        xfbBuffers[buffer].stride = stride;
+        return true;
+    }
+    int addXfbBufferOffset(const TType&);
+    unsigned int computeTypeXfbSize(const TType&, bool& containsDouble) const;
+
 protected:
     void error(TInfoSink& infoSink, const char*);
     void mergeBodies(TInfoSink&, TIntermSequence& globals, const TIntermSequence& unitGlobals);
@@ -191,6 +255,7 @@ protected:
     TIntermNode* treeRoot;
     EProfile profile;
     int version;
+    TBuiltInResource resources;
     int numMains;
     int numErrors;
     bool recursive;
@@ -205,33 +270,12 @@ protected:
     bool pointMode;
     bool xfbMode;
 
-    // for detecting recursion:  pair is <caller, callee>
-    struct TCall {
-        TCall(const TString& pCaller, const TString& pCallee) : caller(pCaller), callee(pCallee) { }
-        TString caller;
-        TString callee;
-        bool visited;
-        bool currentPath;
-        bool errorGiven;
-    };
     typedef std::list<TCall> TGraph;
     TGraph callGraph;
 
-    std::set<TString> ioAccessed;  // set of names of statically read/written I/O that might need extra checking
+    std::set<TString> ioAccessed;       // set of names of statically read/written I/O that might need extra checking
-
-    // A location range is a 2-D rectangle; the set of (location, component) pairs all lying
-    // both within the location range and the component range.
-    // The following are entirely encapsulated by addUsedLocation().
-    struct TRange {
-        int start;
-        int last;
-    };
-    struct TIoRange {
-        TRange location;
-        TRange component;
-        TBasicType basicType;
-    };
     std::vector<TIoRange> usedIo[3];    // sets of used locations, one for each of in, out, and uniform
+    std::vector<TXfbBuffer> xfbBuffers; // all the data we need to track per xfb buffer
 
 private:
     void operator=(TIntermediate&); // prevent assignments